ClawHub
Back to skill

Security audit

ClawdBites

Security checks across malware telemetry and agentic risk

Overview

This skill coherently extracts recipes from public Instagram reels using disclosed media tools, with optional user-directed saving.

Install this only if you are comfortable with the agent downloading public reel media to /tmp, transcribing audio locally, and possibly sending extracted frames to a vision model for reading on-screen recipe text. Use the wishlist or notes options only for recipes you intentionally want saved.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Description-Behavior Mismatch

Medium
Confidence
94% confidence
Finding
The manifest advertises caption-based recipe extraction, but the documented behavior expands into downloading the full video, transcribing audio, and analyzing frames. This is a material scope expansion because it changes the data processed and the operations performed without being clearly reflected in the declared skill purpose, increasing the chance of unexpected collection and processing.

Description-Behavior Mismatch

Medium
Confidence
93% confidence
Finding
The skill is presented as an extraction/display tool, but it also documents saving data to Apple Notes and local JSON storage. That broadens the skill from transient processing into persistent storage and cross-skill data sharing, which can surprise users and create privacy and integrity risks if recipe URLs or user preferences are retained without clear disclosure.

Missing User Warnings

Medium
Confidence
96% confidence
Finding
The workflow instructs the agent to automatically download the reel, extract audio, and transcribe it whenever the caption is incomplete, without a distinct warning or consent checkpoint. This is dangerous because it silently escalates from parsing visible metadata to processing richer media content, which may have different privacy, bandwidth, storage, and copyright implications than the user expected.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The wishlist feature writes recipe data to a persistent local JSON file, but the documentation does not provide an explicit warning at the point of collection that this data will be stored. Persistent storage of source URLs, handles, dates, and recipe metadata can expose user interests and create unintended retention if users believe the interaction is ephemeral.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.